RESEARCH

As a Research Associate at the Data Science Institute, I work on a variety of projects within the health sciences. A summary of my current projects is provided below.

LINKS FOR CURRENT PROJECTS

Predictive Modeling of Sepsis - paper here presented at the 2018 IEEE International Conference on Healthcare Informatics

Dept of Medicine, University of Virginia


Environment & Population Informed Emergency Resource Allocation System (EPIERAS) - project details here

UVA Medical Communications Center & Albemarle Emergency Communications Center


Trauma Acuity Stratification - paper here published in Prehospital Emergency Care

Dept of Surgery, University of Virginia


Computer-Aided Prescribing in Geriatrics - project details here

Dept of Medicine, University of Virginia


Patient-Centric Design for T1D Self-Care - preparing manuscript here for submission to Applied Clinical Informatics

Division of Biomedical Informatics, Dept of Public Health Sciences, University of Virginia



PREDICTIVE MODELING OF SEPSIS

Dept of Medicine, University of Virginia
______________________________________________________________________________________________



Paper
here for our 3 models presented at the 2018 IEEE International Conference on Healthcare Informatics and abstract here for the logistic regression model presented at the 2017 Biomedical Engineering Society Conference

Overview:

I work with the UVa Division of Infectious Disease and International Health to build predictive models of sepsis, seeking to improve the precision and timeliness with which infections can be identified and treated. The models use patient demographics, along with vital signs and blood culture results (pulled from a time window of 4 - 72hrs prior to infection onset), to calculate sepsis risk scores for ICU patients. We test a variety of feature selection methods (e.g., LASSO, RFE, forward selection) and models (logistic regression, random forest, BMA, PCA, SVM) using the Sepsis-3 definition. Ultimately, we strive to integrate the algorithms within the UVa EHR system, allowing for more proficient treatment of sepsis than that achieved with the current approach of using SIRS criteria.

We are currently exploring exciting new avenues of feeding models with predictors such as written physician notes, medications, and therapies. Every week, I meet with UVA physicians who are actively treating patients in the Medical ICU to share new insights and develop new ideas for attacking this elusive pathology. This uniquely intimate interaction allows the research and clinical therapy to build from one another in real-time: a model of research and practice I hope to elevate in prehospital and trauma care.



ENVIRONMENT & POPULATION INFORMED EMERGENCY RESOURCE ALLOCATION SYSTEM (EPIERAS)

UVA Medical Communications Center & Albemarle Emergency Communications Center
______________________________________________________________________________________________



Project details
here

Overview:

Environment & Population Informed Emergency Resource Allocation System (EPIERAS), "Epi," is an algorithm that uses diverse data sources to predict the frequency and nature of local 911 calls, helping us to more efficiently deploy tactical resources and schedule personnel on a daily basis.

Beyond reducing response times and improving the prevention of 911 calls, Epi reveals how social inequities comprise the most potent predictors of both medical and non-medical emergencies. In this way, Epi informs preemptive efforts to address disparities throughout local communities. In addition, these insights allow us to overcome the ignorant, yet pervasive attitudes shaped by narrow perspectives on the mechanisms that lead to elevated 911 calls. That is, Epi forces us to see “bad neighborhoods” or “frequent flyers” not as the product of personal characteristics of those affected, but instead as the product of systemic failures, for which we all must take accountability.



TRAUMA ACUITY STRATIFICATION

Dept of Surgery, University of Virginia
______________________________________________________________________________________________



PubMed link
here and full manuscript here for publication in Prehospital Emergency Care
And paper here under review for publication by American Journal of Emergency Medicine

Overview:

I conduct research with Dr. Jeffrey Young, Professor of Surgery and Director of the Level 1 Trauma Center, and, this past summer, I mentored the UVA medical students who were selected to complete funded projects for the 2018 Medical School Summer Research Program.

I perform the majority of our data analysis for a variety of projects, including those listed above. In our project recently published in Prehospital Emergency Care, I developed a new acuity stratification approach to argue that the growing consensus on one of the biggest questions in trauma, the “golden hour,” is founded on misguided conclusions from studies that made inappropriate assumptions about their trauma populations. The study reveals weaknesses in popular statistical approaches used to evaluate many questions in trauma care and, I contend, has implications relevant to any meta-analysis involving patients of varying acuity.

In continuation of my efforts thus far, I hope to use a career as a physician and researcher to address data limitations by increasing collaboration between dispatch, trauma, and rehabilitation centers to build registries that can be autonomously populated with diverse data types containing comprehensive information from the 911 call, to prehospital and hospital interventions, to full recovery. Such datasets can empower more robust analyses of trauma populations, relieving our current reliance on approaches polluted by a priori assumptions and convoluted imputation methodology that can lead to misguided, dangerous conclusions, as demonstrated in the “golden hour” study.



COMPUTER-AIDED PRESCRIBING IN GERIATRICS

Dept of Medicine, University of Virginia
______________________________________________________________________________________________



Project details
here

Overview:

Beers' Criteria and STOPP-START Criteria were developed to protect against potentially inappropriate prescribing (PIP) in geriatric medicine. However, inefficiencies in usability greatly limit their utility. We sought to address this problem by building an IT system that allows physicians to paste a list of medications into a search bar which then returns the associated recommendations from both Beers’ and STOPP-START Criteria. The system is integrated with the UVA EHR System and can use patient data to alert physicians of PIP. The system is being introduced as a new tool within UVA Internal Medicine to evaluate whether physicians with access to the system will more frequently reference the criteria when prescribing medications to geriatric patients than those without access.



PATIENT-CENTRIC DESIGN FOR T1D SELF-CARE

Division of Biomedical Informatics, Dept of Public Health, Sciences University of Virginia
______________________________________________________________________________________________



Preparing manuscript
here for submission to Applied Clinical Informatics

Overview:

We conducted semi-structured interviews of U.S. students, ages 18-24, with type-1 diabetes to investigate psychosocial challenges of self-care. With the increase in the ubiquity of mobile phones and tablets, this project seeks to evaluate how such devices can be better leveraged in the management of type-1 diabetes.





   
 

_________________________________________________________________________________
_________________________________________________________________________________

phs5eg@virginia.edu